One may obtain an estimate improving your factor of 4 to a factor of 3. 

The ranks of hyperbolic 2-orbifolds were computed by [Zieschang et al][1]. If $\partial\mathcal{O}$ has genus $g$ and $p$ cone points, then they show that $rank(\pi_1\partial\mathcal{O})\leq 2g+p-1$, except in the case $p=0$, one has $rank(\partial\pi_1(\mathcal{O}))=2g$ (of course, one may deduce this estimate
directly by thinking about the punctured case). The same argument (half-lives, half-dies) 
applies in that case (as Igor observes), so I'll assume $p>0$. 


A theorem of Sullivan shows that the deformation space of geometrically finite structures on $\mathcal{O}$ is parameterized by the Teichmuller space of $\partial{\mathcal{O}}$. This follows from the theory of quasiconformal deformations of Kleinian groups. Now, one follows the proof of the Ahlfors finiteness theorem. If $rank(\pi_1\mathcal{O})=k$, then the space of deformations of representations of $\pi_1\mathcal{O}$ into $PSL_2(\mathbb{C})$ up to conjugacy has $\mathbb{C}$-dimension $\leq 3k-3$ (this follows by computing the dimension of the variety of representations, and using that the conjugacy action is faithful since the generators are non-commuting). This is also the dimension of the space of geometrically finite reps., since these are structurally stable. 

On the other hand, the Teichmuller space of $\partial\mathcal{O}$ has complex dimension $3g-3+p$, so we get $3g-3+p\leq 3k-3$, or $g+p/3\leq k$. From the rank computation above, then we get $\frac13 rank(\pi_1\partial\mathcal{O})\leq  2g/3+p/3-1/3 \leq g+p/3 \leq k$. Obviously the worst estimate holds when $g=0$. One might be able to improve this result taking into account the relators. 


  [1]: http://www.ams.org/mathscinet-getitem?mr=382457